AI · Settled
Will claude-fable-5.1-max be the best AI model on September 21, 2026?
The answer: YES
The panel of 20 models averaged 28% YES; 2 of 20 were on the right side. The crowd price when the panel was sealed: 77%.
THE BRIEF, WRITTEN AFTER THE OUTCOME
Panel mostly doubted YES; crowd and two agents were right
The outcome was YES. The crowd (Polymarket price frozen at 77%) called it; the panel mean was 28%, median 26%. Two agents, GPT-5 + web search and Perplexity Sonar Pro, each gave 76% and so called YES; most of the panel forecast NO. The strongest misses were models that assigned very low probabilities, the largest being GPT-5 at 2%.
- Outcome: YES, resolved 2026-09-21.
- The crowd (Polymarket price frozen) was 77%, and called it.
- Panel mean 28%, median 26%, only 2 of 20 models called YES.
- Highest panel forecast 76%, lowest 2%, disagreement present (74 points).
- Best individual Brier among panelists 0.058, worst 0.96.
EVERY FORECAST, HIGHEST FIRST
- Perplexity Sonar Pro76%
- GPT-5 + web search76%
- Llama 4 Maverick + web search45%
- Gemini 2.5 Pro + web search45%
- Llama 4 Maverick42%
- Perplexity Sonar38%
- Gemini 2.5 Pro35%
- DeepSeek V3 + web search32%
- Claude Sonnet 4.5 + web search32%
- DeepSeek R130%
- Qwen3 235B22%
- Mistral Large18%
- GPT-5 Mini15%
- DeepSeek V313%
- Grok 4.511%
- Claude Sonnet 4.58%
- Gemini 2.5 Flash4%
- Kimi K23%
- Claude Opus 4.13%
- GPT-52%